THz Radar Cross Section(RCS)measurement setup based on THz Time Domain Spectroscopy(TDS)is built to provide large scaled targets test ability in recent years.As calibrations,the metal plates and dihedrons are used in ...THz Radar Cross Section(RCS)measurement setup based on THz Time Domain Spectroscopy(TDS)is built to provide large scaled targets test ability in recent years.As calibrations,the metal plates and dihedrons are used in our experiments.The measurements are performed in a monostatic terahertz time-domain setup.The author proposed time domain and frequency domain calibration methods for angular RCS of calibrations,comparing the measurements with the theory to verify the ability of the time domain measurement setup.展开更多
在跨场景分类任务中,大多数领域自适应方法(Domain Adaptation,DA)关注于源域数据和目标域数据由相同传感器获得且具有相同地物类别的迁移任务,然而当目标数据中存在新类别时自适应性能会显著下降.此外,大多数高光谱图像分类方法采用全...在跨场景分类任务中,大多数领域自适应方法(Domain Adaptation,DA)关注于源域数据和目标域数据由相同传感器获得且具有相同地物类别的迁移任务,然而当目标数据中存在新类别时自适应性能会显著下降.此外,大多数高光谱图像分类方法采用全局表征机制,即针对固定大小窗口的样本进行表征学习,其地物类别表征能力有限.本文提出了一种基于局部表征的少样本学习框架(Local representation Few Shot Learning,LrFSL),尝试在少样本学习中构建局部表征机制突破全局表征能力上限.在提出框架中,对所有具有标签的源域数据和少量具有标签的目标域数据构建元任务,依照元学习策略同步进行情景训练,与此同时设计了域内局部表征模块(Intra-domain Local Representation block,ILR-block)用于挖掘样本中多个局部表征的语义信息,设计了域间局部对齐模块(Inter-domain Local Alignment block,ILA-block)进行跨域逐类别分布对齐以缓解领域偏移对少样本学习的影响.在三个公开高光谱图像数据集上的实验结果证明了该方法显著优于目前最先进的方法.展开更多
A conformal multi-resolution time-domain( CMRTD) method is presented for modeling curved objects. The effective dielectric constant and area weighting are used to derive the update equations of CMRTD. The backward sca...A conformal multi-resolution time-domain( CMRTD) method is presented for modeling curved objects. The effective dielectric constant and area weighting are used to derive the update equations of CMRTD. The backward scattering bistatic radar cross sections( RCS) of the dielectric cylinder and ellipsoid are used to validate the proposed method. The results show that the proposed conformal method is more accurate to deal with the complex curved objects in electromagnetic simulations.展开更多
Several major challenges need to be faced for efficient transient multiscale electromagnetic simulations, such as flex- ible and robust geometric modeling schemes, efficient and stable time-stepping algorithms, etc. F...Several major challenges need to be faced for efficient transient multiscale electromagnetic simulations, such as flex- ible and robust geometric modeling schemes, efficient and stable time-stepping algorithms, etc. Fortunately, because of the versatile choices of spatial discretization and temporal integration, a discontinuous Galerkin time-domain （DGTD） method can be a very promising method of solving transient multiscale electromagnetic problems. In this paper, we present the application of a leap-frog DGTD method to the analyzing of the multiscale electromagnetic scattering problems. The uniaxial perfect matching layer （UPML） truncation of the computational domain is discussed and formulated in the leap-frog DGTD context. Numerical validations are performed in the challenging test cases demonstrating the accuracy and effectiveness of the method in solving transient multiscale electromagnetic problems compared with those of other numerical methods.展开更多
基金The Science and Technology Commission of Shanghai Municipality under Grant(16ZR1435000)
文摘THz Radar Cross Section(RCS)measurement setup based on THz Time Domain Spectroscopy(TDS)is built to provide large scaled targets test ability in recent years.As calibrations,the metal plates and dihedrons are used in our experiments.The measurements are performed in a monostatic terahertz time-domain setup.The author proposed time domain and frequency domain calibration methods for angular RCS of calibrations,comparing the measurements with the theory to verify the ability of the time domain measurement setup.
文摘在跨场景分类任务中,大多数领域自适应方法(Domain Adaptation,DA)关注于源域数据和目标域数据由相同传感器获得且具有相同地物类别的迁移任务,然而当目标数据中存在新类别时自适应性能会显著下降.此外,大多数高光谱图像分类方法采用全局表征机制,即针对固定大小窗口的样本进行表征学习,其地物类别表征能力有限.本文提出了一种基于局部表征的少样本学习框架(Local representation Few Shot Learning,LrFSL),尝试在少样本学习中构建局部表征机制突破全局表征能力上限.在提出框架中,对所有具有标签的源域数据和少量具有标签的目标域数据构建元任务,依照元学习策略同步进行情景训练,与此同时设计了域内局部表征模块(Intra-domain Local Representation block,ILR-block)用于挖掘样本中多个局部表征的语义信息,设计了域间局部对齐模块(Inter-domain Local Alignment block,ILA-block)进行跨域逐类别分布对齐以缓解领域偏移对少样本学习的影响.在三个公开高光谱图像数据集上的实验结果证明了该方法显著优于目前最先进的方法.
基金Supported by the National Natural Science Foundation of China(61172024)the Funding of Jiangsu Innovation Program for Graduate Education and the Fundamental Research Funds for the Central Universities(CXZZ12-0156)
文摘A conformal multi-resolution time-domain( CMRTD) method is presented for modeling curved objects. The effective dielectric constant and area weighting are used to derive the update equations of CMRTD. The backward scattering bistatic radar cross sections( RCS) of the dielectric cylinder and ellipsoid are used to validate the proposed method. The results show that the proposed conformal method is more accurate to deal with the complex curved objects in electromagnetic simulations.
基金supported by the National Natural Science Foundation of China(Grant Nos.61301056 and 11176007)the Sichuan Provincial Science and Technology Support Program,China(Grant No.2013HH0047)+1 种基金the Fok Ying Tung Education Foundation,China(Grant No.141062)the"111"Project,China(Grant No.B07046)
文摘Several major challenges need to be faced for efficient transient multiscale electromagnetic simulations, such as flex- ible and robust geometric modeling schemes, efficient and stable time-stepping algorithms, etc. Fortunately, because of the versatile choices of spatial discretization and temporal integration, a discontinuous Galerkin time-domain （DGTD） method can be a very promising method of solving transient multiscale electromagnetic problems. In this paper, we present the application of a leap-frog DGTD method to the analyzing of the multiscale electromagnetic scattering problems. The uniaxial perfect matching layer （UPML） truncation of the computational domain is discussed and formulated in the leap-frog DGTD context. Numerical validations are performed in the challenging test cases demonstrating the accuracy and effectiveness of the method in solving transient multiscale electromagnetic problems compared with those of other numerical methods.
